Week 8B: Raster Data in the Wild¶
Oct 22, 2020
This week: raster data analysis with the holoviz ecosystem¶
Two case studies:
- Last time: Using satellite imagery to detect changes in lake volume
- Today: Detecting urban heat islands in Philadelphia
Initialize our packages:
In [4]:
from matplotlib import pyplot as plt
import numpy as np
import pandas as pd
import geopandas as gpd
%matplotlib inline
In [6]:
import intake
import xarray as xr
import cartopy.crs as ccrs
In [7]:
import hvplot.xarray
import hvplot.pandas
import holoviews as hv
import geoviews as gv
from geoviews.tile_sources import EsriImagery
hv.extension('bokeh')
Create the intake data catalog so we can load our datasets:
In [8]:
url = 'https://raw.githubusercontent.com/pyviz-topics/EarthML/master/examples/catalog.yml'
cat = intake.open_catalog(url)
Example #2: Urban heat islands¶
- We'll reproduce an analysis by Ken on urban heat islands in Philadelphia using Python.
- The analysis uses Landsat 8 data (2017)
- See: http://urbanspatialanalysis.com/urban-heat-islands-street-trees-in-philadelphia/
First load some metadata for Landsat 8¶
In [9]:
band_info = pd.DataFrame([
(1, "Aerosol", " 0.43 - 0.45", 0.440, "30", "Coastal aerosol"),
(2, "Blue", " 0.45 - 0.51", 0.480, "30", "Blue"),
(3, "Green", " 0.53 - 0.59", 0.560, "30", "Green"),
(4, "Red", " 0.64 - 0.67", 0.655, "30", "Red"),
(5, "NIR", " 0.85 - 0.88", 0.865, "30", "Near Infrared (NIR)"),
(6, "SWIR1", " 1.57 - 1.65", 1.610, "30", "Shortwave Infrared (SWIR) 1"),
(7, "SWIR2", " 2.11 - 2.29", 2.200, "30", "Shortwave Infrared (SWIR) 2"),
(8, "Panc", " 0.50 - 0.68", 0.590, "15", "Panchromatic"),
(9, "Cirrus", " 1.36 - 1.38", 1.370, "30", "Cirrus"),
(10, "TIRS1", "10.60 - 11.19", 10.895, "100 * (30)", "Thermal Infrared (TIRS) 1"),
(11, "TIRS2", "11.50 - 12.51", 12.005, "100 * (30)", "Thermal Infrared (TIRS) 2")],
columns=['Band', 'Name', 'Wavelength Range (µm)', 'Nominal Wavelength (µm)', 'Resolution (m)', 'Description']).set_index(["Band"])
band_info
Out[9]:
Landsat data on Google Cloud Storage¶
We'll be downloading publicly available Landsat data from Google Cloud Storage
See: https://cloud.google.com/storage/docs/public-datasets/landsat
The relevant information is stored in our intake catalog:
From our catalog.yml file:
google_landsat_band:
description: Landsat bands from Google Cloud Storage
driver: rasterio
parameters:
path:
description: landsat path
type: int
row:
description: landsat row
type: int
product_id:
description: landsat file id
type: str
band:
description: band
type: int
args:
urlpath: https://storage.googleapis.com/gcp-public-data-landsat/LC08/01/{{ '%03d' % path }}/{{ '%03d' % row }}/{{ product_id }}/{{ product_id }}_B{{ band }}.TIF
chunks:
band: 1
x: 256
y: 256From the "urlpath" above, you can see we need "path", "row", and "product_id" variables to properly identify a Landsat image:
The path and row corresponding to the area over Philadelphia has already been selected using the Earth Explorer. This tool was also used to find the id of the particular date of interest using the same tool.
In [10]:
# Necessary variables
path = 14
row = 32
product_id = 'LC08_L1TP_014032_20160727_20170222_01_T1'
Use a utility function to query the API and request a specific band¶
This will return a specific Landsat band as a dask array.
In [11]:
from random import random
from time import sleep
def get_band_with_exponential_backoff(path, row, product_id, band, maximum_backoff=32):
"""
Google Cloud Storage recommends using exponential backoff
when accessing the API.
https://cloud.google.com/storage/docs/exponential-backoff
"""
n = backoff = 0
while backoff < maximum_backoff:
try:
return cat.google_landsat_band(
path=path, row=row, product_id=product_id, band=band
).to_dask()
except:
backoff = min(2 ** n + random(), maximum_backoff)
sleep(backoff)
n += 1
Load all of the bands and combine them into a single xarray DataArray¶
Loop over each band, load that band using the above function, and then concatenate the results together..
In [12]:
bands = [1, 2, 3, 4, 5, 6, 7, 9, 10, 11]
datasets = []
for band in bands:
da = get_band_with_exponential_backoff(
path=path, row=row, product_id=product_id, band=band
)
da = da.assign_coords(band=[band])
datasets.append(da)
ds = xr.concat(datasets, "band", compat="identical")
ds
Out[12]:
Important: CRS info¶
CRS is EPSG=32618
Also grab the metadata from Google Cloud Storage¶
- There is an associated metadata file stored on Google Cloud Storage
- The below function will parse that metadata file for a specific path, row, and product ID
- The specifics of this are not overly important for our purposes
In [13]:
def load_google_landsat_metadata(path, row, product_id):
"""
Load Landsat metadata for path, row, product_id from Google Cloud Storage
"""
def parse_type(x):
try:
return eval(x)
except:
return x
baseurl = "https://storage.googleapis.com/gcp-public-data-landsat/LC08/01"
suffix = f"{path:03d}/{row:03d}/{product_id}/{product_id}_MTL.txt"
df = intake.open_csv(
urlpath=f"{baseurl}/{suffix}",
csv_kwargs={
"sep": "=",
"header": None,
"names": ["index", "value"],
"skiprows": 2,
"converters": {"index": (lambda x: x.strip()), "value": parse_type},
},
).read()
metadata = df.set_index("index")["value"]
return metadata
In [14]:
metadata = load_google_landsat_metadata(path, row, product_id)
metadata.head()
Out[14]:
Let's trim our data to the Philadelphia limits¶
- The Landsat image covers an area much wider than the Philadelphia limits
- We'll load the city limits from Open Data Philly
- Use the
slice()function discussed last lecture
Load the city limits¶
- From OpenDataPhilly, the city limits for Philadelphia are available at: http://data.phl.opendata.arcgis.com/datasets/405ec3da942d4e20869d4e1449a2be48_0.geojson
- Be sure to convert to the same CRS as the Landsat data!
In [17]:
# Load the GeoJSON from the URL
url = "http://data.phl.opendata.arcgis.com/datasets/405ec3da942d4e20869d4e1449a2be48_0.geojson"
city_limits = gpd.read_file(url)
# Convert to the right CRS for this data
city_limits = city_limits.to_crs(epsg=32618)
Use the xmin/xmax and ymin/ymax of the city limits to trim the Landsat DataArray¶
- Use the built-in slice functionality of xarray
- Remember, the
ycoordinates are in descending order, so you'll slice will need to be in descending order as well
In [24]:
# Look up the order of xmin/xmax and ymin/ymax
city_limits.total_bounds?
In [18]:
# Get x/y range of city limits from "total_bounds"
xmin, ymin, xmax, ymax = city_limits.total_bounds
In [19]:
# Slice our xarray data
subset = ds.sel(
x=slice(xmin, xmax),
y=slice(ymax, ymin)
) # NOTE: y coordinate system is in descending order!
Call the compute() function on your sliced data¶
- Originally, the Landsat data was stored as dask arrays
- This should now only load the data into memory that covers Philadelphia
In [20]:
subset = subset.compute()
subset
Out[20]:
In [21]:
# Original data was about 8000 x 8000 in size
ds.shape
Out[21]:
In [22]:
# Sliced data is only about 1000 x 1000 in size!
subset.shape
Out[22]:
Let's plot band 3 of the Landsat data¶
In [23]:
# City limits plot
limits_plot = city_limits.hvplot.polygons(
line_color="white", alpha=0, line_alpha=1, crs=32618, geo=True
)
# Landsat plot
landsat_plot = subset.sel(band=3).hvplot.image(
x="x",
y="y",
rasterize=True,
cmap="viridis",
frame_width=500,
frame_height=500,
geo=True,
crs=32618,
)
# Combined
landsat_plot * limits_plot
Out[23]:
Calculate the NDVI¶
Remember, NDVI = (NIR - Red) / (NIR + Red)
In [24]:
band_info.head()
Out[24]:
NIR is band 5 and Red is band 4
In [25]:
# Selet the bands we need
NIR = subset.sel(band=5)
RED = subset.sel(band=4)
# Calculate the NDVI
NDVI = (NIR - RED) / (NIR + RED)
NDVI = NDVI.where(NDVI < np.inf)
NDVI.head()
Out[25]:
In [26]:
# NEW: Use datashader to plot the NDVI
p = NDVI.hvplot.image(
x="x",
y="y",
geo=True,
crs=32618,
datashade=True, # NEW: USE DATASHADER
project=True, # NEW: PROJECT THE DATA BEFORE PLOTTING
frame_height=500,
frame_width=500,
cmap="viridis",
)
# NEW: Use a transparent rasterized version of the plot for hover text
# No hover text available with datashaded images so we can use the rasterized version
p_hover = NDVI.hvplot(
x="x",
y="y",
crs=32618,
geo=True,
rasterize=True,
frame_height=500,
frame_width=500,
alpha=0, # SET ALPHA=0 TO MAKE THIS TRANSPARENT
colorbar=False,
)
# City limits
limits_plot = city_limits.hvplot.polygons(
geo=True, crs=32618, line_color="white", line_width=2, alpha=0, line_alpha=1
)
p * p_hover * limits_plot
Out[26]:
A couple of notes¶
- Notice that we leveraged datashader algorithm to plot the NDVI by specifying the
datashade=Truekeyword - Hover tools won't work with datashaded images — instead, we overlaid a transparent rasterized version of the image, which shows the mean pixel values across the image
- We'll also need to specify one more for transforming satellite temperature (brightness temp) to land surface temperature.
- For these calculations we'll use both Thermal Infrared bands - 10 and 11.
In [27]:
from rio_toa import brightness_temp, toa_utils
In [28]:
def land_surface_temp(BT, w, NDVI):
"""Calculate land surface temperature of Landsat 8
temp = BT/1 + w * (BT /p) * ln(e)
BT = At Satellite temperature (brightness)
w = wavelength of emitted radiance (μm)
where p = h * c / s (1.439e-2 mK)
h = Planck's constant (Js)
s = Boltzmann constant (J/K)
c = velocity of light (m/s)
"""
h = 6.626e-34
s = 1.38e-23
c = 2.998e8
p = (h * c / s) * 1e6
Pv = (NDVI - NDVI.min() / NDVI.max() - NDVI.min()) ** 2
e = 0.004 * Pv + 0.986
return BT / 1 + w * (BT / p) * np.log(e)
In [30]:
def land_surface_temp_for_band(band):
"""
Utility function to get land surface temperature for a specific band
"""
# params from general Landsat info
w = band_info.loc[band]["Nominal Wavelength (µm)"]
# params from specific Landsat data text file for these images
ML = metadata[f"RADIANCE_MULT_BAND_{band}"]
AL = metadata[f"RADIANCE_ADD_BAND_{band}"]
K1 = metadata[f"K1_CONSTANT_BAND_{band}"]
K2 = metadata[f"K2_CONSTANT_BAND_{band}"]
at_satellite_temp = brightness_temp.brightness_temp(
subset.sel(band=band).values, ML, AL, K1, K2
)
return land_surface_temp(at_satellite_temp, w, NDVI)
Get land surface temp for band 10¶
In [31]:
# temperature for band 10
band = 10
temp_10 = land_surface_temp_for_band(band).compute()
# convert to Farenheit
temp_10_f = toa_utils.temp_rescale(temp_10, 'F')
# Save the numpy array as an xarray
band_10 = subset.sel(band=band).copy(data=temp_10_f)
Get the land surface temp for band 11¶
In [32]:
# Get the land surface temp
band = 11
temp_11 = land_surface_temp_for_band(band).compute()
# Convert to Farenheit
temp_11_f = toa_utils.temp_rescale(temp_11, "F")
# Save as an xarray DataArray
band_11 = subset.sel(band=band).copy(data=temp_11_f)
Plot both temperatures side-by-side¶
In [33]:
# plot for band 10
plot_10 = band_10.hvplot.image(
x="x",
y="y",
rasterize=True,
cmap="fire_r",
crs=32618,
geo=True,
frame_height=400,
frame_width=400,
title="Band 10",
)
# plot for band 11
plot_11 = band_11.hvplot.image(
x="x",
y="y",
rasterize=True,
cmap="fire_r",
crs=32618,
geo=True,
frame_height=400,
frame_width=400,
title="Band 11",
)
plot_10 + plot_11
Out[33]:
Plot the final product (the average of the bands)¶
In [34]:
# Let's average the results of these two bands
mean_temp_f = (band_10 + band_11) / 2
# IMPORTANT: copy over the metadata
mean_temp_f.attrs = band_10.attrs
In [35]:
mean_temp_f
Out[35]:
In [36]:
p = mean_temp_f.hvplot.image(
x="x",
y="y",
rasterize=True,
crs=32618,
geo=True,
frame_width=600,
frame_height=600,
cmap="rainbow",
alpha=0.5,
legend=False,
title="Mean Surface Temp (F)"
)
img = p * EsriImagery
img
Out[36]:
Use the zoom tool to identify hot spots¶
Key insight: color of the building roof plays a very important role in urban heat islands
Exercise: isolate hot spots on the map¶
We can use the .where() function in xarray to identify the heat islands across the city
To do:
- Select only those pixels with mean temperatures about 90 degrees.
- Remember, the
where()function takes a boolean array where each pixel is True/False based on the selection criteria - Use hvplot to plot the imagery for Philadelphia, with the hotspots overlaid
In [ ]:
What about cold spots?¶
Let's select things less than 75 degrees as well...
In [ ]:
Exercise: Exploring the relationship between temperature and parks/trees¶
In this section, we'll use zonal statistics to calculate the average temperatures for neighborhoods and parks, and compare this to the number of street trees in the city.
Saving xarray data to a geoTIFF file¶
- Unfortunately, the zonal statistics calculation requires a rasterio file — so, we will need to to save our mean temperature xarray DataArray to a
.tiffile - This is not built in to xarray (yet), but we can use the rioxarray package
In [41]:
import rioxarray
In [42]:
# Save to a raster GeoTIFF file!
mean_temp_f.rio.to_raster("./mean_temp_f.tif")
Part 1.1: Calculate the mean temperature for parks in Philadelphia¶
- The parks dataset is available in the data folder:
./data/parks.geojson - Use the
rasterstatspackage to calculate the zonal stats - Look back at week 4's lecture for help!
- Make sure the CRS matches!
In [43]:
from rasterstats import zonal_stats
import rasterio as rio
In [ ]:
Part 1.2: Plot a histogram of temperature values for each park¶
- Before plotting the histogram, subtract out the mean temperature of the entire city
- This will quickly let us see which parks have average temperatures above or below the citywide average
In [ ]:
In [ ]:
Part 2.1: Calculate the average temperature in Philadelphia's neighborhoods¶
- Neighborhoods can be loaded from: https://github.com/azavea/geo-data/raw/master/Neighborhoods_Philadelphia/Neighborhoods_Philadelphia.geojson
- Again, use zonal statistics to do the calculation
- Make sure the CRS matches!
In [ ]:
Part 2.2: Plot a choropleth map of the neighborhoods, coloring by temperature¶
- Make a static chart with Geopandas
- Make an interactive chart with Altair / hvplot, so we can quickly identify the neighborhoods with the highest/lowest temperatures!
Plot using geopandas¶
In [ ]:
Plot using altair/hvplot¶
In [ ]:
Part 2.3: Calculate the number of street trees per neighborhood¶
- We can use the street tree inventory from OpenDataPhilly
- The tree data can be loaded from: http://data.phl.opendata.arcgis.com/datasets/957f032f9c874327a1ad800abd887d17_0.geojson
- You will need to use the
sjoin()function from geopandas to join the tree dataset to the neighborhoods dataset and then groupby the neighborhoods to find the number of trees per neighborhood
In [ ]:
In [ ]:
Part 2.4: Use seaborn to explore the relationship b/w street trees and temperature¶
- Use the
kdeplot()function to plot the relationship between the number of trees per neighborhood and the mean temperature per neighborhood - You will need to
merge()the relevant data frames together based on the neighborhood name
What are the major trends? Do you see the relationship you expect?"
In [73]:
import seaborn as sns
In [ ]: